Hacker News new | ask | show | jobs
by knadh 2237 days ago
Hey, sorry about that. And we're always sorry about technical issues when they happen, and 9/10 times, they stem from systems outside of our control.

The underlying infrastructure in the Indian capital markets is very constrained and was not really built for the kind of scale we're operating at in 2020. The next biggest broker processes only a fraction of our daily volumes. So it really is a matter of scale, industry wide.

To give you an example, the leased lines NSE (National Stock Exchange) offers to brokers to send orders take up to four months to commission, and each line has a max capacity of 1000 messages / second. We've been processing close to 8 million trades a day. So you can imagine how painful things can be.

We've been working hard over the years to reduce legacy market infrastructure dependencies and have been making significant progress.

3 comments

Wait, so what's the problem here?

You're processing an average of 1.33 million trades/hour. Let's double that and round up for peak traffic, that's 3 million trades/hour or 833.33 trades/second.

Unless I'm missing something, you only need 1 line. What's the problem here?

That makes it sound like the only thing involved in stock broking setup is pushing messages through a line :) Market volatility can cause several hours worth of orders to come through in a single burst. What goes on inside a stock broker is actually incredibly complex. I guess that warrants a long blog post.
You bet, it does. With a case study would be interesting to read. At least that is worth for all the loses on that 1/10 because of your technical issues.
Saying sorry doesn't fix the losses. Pro traders are used to taking losses and can manage it. But it is super bad when we lose more than 50% of a huge profit in case of a exceptional market move just because of Kite not able to handle the load and accept an order. Or worst when there is an exceptional loss that just doubles because of system failure. You guys did a great job on the platform, but you should provide reliability to professionals who don't mind paying you extra for that
Thanks, but paying extra for reliability implies that reliability is being compromised and can be provided for a fee, which isn't the case here. As unfortunate as these incidents are, they are rooted in the legacy infrastructure a stock broker is dependent upon and we've been working hard over the years replacing those, component by component.
You became the largest broker, next may be you could become an exchange :D
What exactly are the legacy market infra dependencies and what would you replace them with? 8 million trades a day with 1000 messages / second doesn't sound too challenging. But it is besides the point w.r.t the user experience complaints people are raising here.
You can't send more than x orders in same line unlike traditional network where you can saturate the bandwidth.